1 Introduction 2 the Principles FATI and FITA

1 Introduction 2 the Principles FATI and FITA

View metadata, citation and similar papers at core.ac.uk brought to you by CORE provided by Eldorado - Ressourcen aus und für Lehre, Studium und Forschung Relational and Extensional Interpretation of FATI and FITA Karl-Heinz Temme Abstract A common way to interpret fuzzy IF-THEN rule bases is based on an extensional interpretation, either with a relational approach or without any other constraints. In this paper we investigate both alternatives in combination with the two general paradigms FATI and FITA to process rule bases. Keywords: fuzzy IF-THEN rule bases, relational interpretation, extensional interpretation, FATI, FITA 1 Introduction When interpreting Fuzzy IF-THEN rules one can look at one rule defining a fuzzy relation based on the 0 left and right hand side of the rule. Given a fuzzy input set F the evaluation of this input by a fuzzy 0 rule can be interpreted as the fuzzy image of F under this relation. When interpreting rule bases, there are two principles to perform the above interpretation on more than one rule. The property, whether the relation solves a particular equation system or not, together with the two principles leads to a combination of four alternatives which are now investigated. 2 The Principles FATI and FITA The principles FATI (First Aggregation Then Inference) and FITA (First Inference Then Aggregation) have been defined and discussed deeply in [LRTT98]. FATI and FITA are paradigms to interpret Fuzzy IF-THEN rules and rule bases. When interpreting a rule base according to the FATI principle first one single interpretation I is aggre- 0 0 F gated based on all rules of the rule base. After that the output set G is infered from the input set by means of I in one inference step. I When interpreting a rule base according to the FITA principle first for each rule an interpretation 1 to 0 0 0 I G G F I I 1 n n is generated. Then intermediate sets to are infered from the input set by means of to n 1 0 0 0 G G G in n parallel inference steps. Finally the output set is aggregated from to . n 1 For both paradigms the aggregation resp. generation of the interpretation and/or the aggregation of the output set can be performed as well in an extensional as in a non-extensional way (see below). The following functional operators are based only on extensional definitions, non-extensional approaches will be discussed in forthcoming papers. Furthermore the interpretations are restricted to relational-based ones. 1 3 Prerequisites and Definitions Let n 2N 0 0 0 0 F; F ;G;G ;F ;F ;G ;G : U !h0; 1i i =1;::: ;n U i i be fuzzy sets over , i i 2 : h 0; 1i !h0; 1i ; fuzzy connectives, S; S : U U !h0; 1i i =1;::: ;n i fuzzy relations, u; v ; x; y 2 U IF F THEN G i =1;::: ;n i i a fuzzy rule base, n : h0; 1i !h0; 1i n ; -ary functions over the unit intervall. F F S Definition S is the image of under , defined as: S F )(y )= Sup f (F (x);S(x; y )) x 2 U g y 2 V ( def Various concepts of correctness are discussed in [LRTT98]. A functional operator works locally correct I RB I with a given rule base RB and a given interpretation ,if is rule-wise correct with respect to according to definition 3.2.2 on page 27 in [LRTT98]. 4 The Functional Operator FATI According to the principle FATI first one interpretation is generated by aggregating individual interpre- tations, here based on fuzzy relations and due to the extensional approach in two steps: S (x; y )= (F (x);G (y )) i =1;::: ;n i i 1. i def S (x; y )= (S (x; y );S (x; y );::: ;S (x; y )) 2 n 2. def 1 0 0 S F Finally the inference is performed based on the input set F and the interpretation as the image of under S . ATI Definition Functional operator F 0 0 ATI (F )(y )= Sup (F (x);S(x; y )) x 2 U F def 5 The Functional Operator FITA According to the principle FITA first n (extensional) interpretations are generated: S (x; y )= (F (x);G (y )) i =1;::: ;n i i i def , 0 n F S Then inferences are performed as images of under the corresponding i : 0 0 G (y ) = Sup (F (x);S (x; y )) x 2 U i =1;::: ;n i , i 0 Finally the output set G is aggregated: 0 0 0 0 G (y )= (G (y );G (y );::: ;G (y )) 1 2 n 2 A Definition Functional operator FIT 0 A(F )(y )= FIT def 0 0 (Sup f(F (x);S (x; y )) x 2 U g ;::: ;Sup f(F (x);S (x; y )) x 2 U g ) 1 n 6 Relational and Extensional Interpretations FITA The functional operators FATI and are defined by means of relations over fuzzy sets. If these y F (x) G(y ) relations do not directly depend on x and but on the membership values and we call them “defined in an extensional way” and call functional operators based on them “extensional interpreta- tions”. A relational interpretation is: 1. an extensional interpretation S (x; y ) 2. for a given aggregation function the relation has to be a solution of the relational equation system If for a given rule base such a solution does not exist, then a relation interpretration is not possible. 7 The Functional Operator FATI with Relational Rule Interpretation The relational rule interpretation requires, that the relation S is a solution of the equation system: S F = G 1 1 S F = G 2 2 . S F = G n n In general this puts certain constraints on the selection of and . For particular rule bases there may even exist no solution. The relational approach guarantees by definition, that the FATI operator works locally correct [Thiel95], FATI(F )=G i =1;::: ;n i e. g. i holds for all . 8 The Functional Operator FATI with Extensional Rule Interpretation When using an extensional rule interpretation, und can be chosen arbitrarily. This includes the above relational approach but what about the other cases ? One can distinguish two alternatives: S (8i : S F = G ) 9j : S F 6= G i i i j j Case 1: All i are local solutions but S F 6= G j j j Case 2: j for at least one FATI F G i For both cases at least one point exists, in which a operator based on the i and from above F FATI(F ) 6= G j j does not work locally correctly, e. g. a j exists with . 3 9 The Functional Operator FITA with Relational Rule Interpretation Again the relational rule interpretation requires, that the relation S solves the equation system: S F = G 1 1 S F = G 2 2 . S F = G n n FITA FITA As the relation S is not used in the definition of the operator, a relational interpretation for FITA S is senseless. One can weaken a relational interpretation for in a way that only local solutions i have to exist, but even this weakened approach does not guarantee local correctness as for arbitrary the S FITA local correctness of the i is not carried over to the operator . S One can force local correctness if for example the following constraints are put on the i and : 8i8y : (0; :::; 0;G (y ); 0; :::; 0) = G (y ) i i and 8i : S F = ; j 6= i j i for n F Example: is a -ary extension of a S-norm and all i are pairwise disjoint. 10 The Functional Operator FITA with Extensional Rule Interpretation When using an extensional rule interpretation, und can be chosen arbitrarily. This includes the F above (weakened) relational approach but in addition all cases for which at least one j exists with S F 6= G j j j . In general local correctness is not guaranteed. 11 Conclusions ATI Among the four alternatives only one approach, the operator F with relational rule interpretation, A guarantees the correctness of the interpretation. For FIT the relational approach is senseless, maybe the property of local correctness supports the definition of an operator for interpretation a little bit. In all other three cases one has to discuss what it means that a functional operator for the interpretation of a fuzzy rule base does not yield the fuzzy set on the right side of a rule if fed by the fuzzy set of the left side of the rule. References [Thiel95] H. THIELE: On the Calculus of Fuzzy IF-THEN Rules, in: H.-J. Zimmermann, M. G. Ne- goita, D. Dascalu (eds): Real World Applications of Intelligent Technologies, Editura Academiei Romane, Bucharest, Romania, 1995. [LRTT98] S. LEHMKE,B.REUSCH, K.-H. TEMME,H.THIELE: On Interpreting Fuzzy IF-THEN Rule Bases by Concepts of Functional Analysis, Technical Report CI-19/98, ISSN 1433- 3325, Collaborative Research Center 531, University of Dortmund, Germany, February 1998. 4 Example for FATI with relational rule interpretation IF F THEN G 1 1 IF F THEN G 2 2 =min y =max G1 1 G2 F1 F2 0 x ( a =1 b if (a; b)= S (x; y ) = max ( (F (x);G (y ));(F (x);G (y ))) 1 1 2 2 0 otherwise y 1 0 x x1 x2 S is a solution of the equation system S F = G 1 1 S F = G 2 2 FATI F The operator defined by , , and works correctly, here shown for the input 1 .

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    8 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us